Welcome back to the fifth and final session of this year's digital conference. This is our live manager Q&A. Hopefully, you've enjoyed the lineup so far. We've already heard from Professor Hannah Fry and Tom talking about signals from the future. We heard from Linda, Tom, and Lawrence on China and the opportunities there for Scottish Mortgage. And we just heard from Ethan Mollick and Lawrence on the next phase of AI. And aren't there so many interesting questions in that subject, like what does AI mean for markets, for industries, for companies? And I think everyone's sort of grappling with one version of a common question, which is what does AI mean for me, for my life, for my role, for my family, and for my investments. Now, clearly no firm answers, but hopefully that session had some interesting lines of inquiry.
Now, as promised, we have your Scottish Mortgage managers in the hot seat ready to answer your questions. And goodness me, we've had a lot of questions. Thank you so much for engaging so strongly with this. The SM audience is always super curious and engaged and you have not disappointed on this occasion. There are so many questions in fact that we'll not get through all of them. But I think if we're smart and we approach the sort of big themes in the questions, we should be able to satisfy the larger part of our audience's curiosity. So without further ado, let's kick off. And Tom, I'd like to start with SpaceX. No surprise, it was one of the biggest themes within the questions. And one of the questions was, it's about the position size. It's grown to an unusually large holding. How did we get here? And how are you thinking about that concentration risk and that liquidity in terms of the windows?
Sure. So SpaceX has been a phenomenally successful investment for us. But I think you know that structure of returns is one that I talked a little bit about with Hannah. You shouldn't be surprised by that. You know, your returns are driven by a small number of big winners and so our job is to try and maximize the return from those big winners for our shareholders. And that means positions will get lodged that, you know, if you sell prematurely, you know, you might keep the position sizes even, which, you know, everybody's happy because there isn't that low risk. Well, I don't, we don't equate large positions with risk. It's more about can you run your big winners and make sure you capture all of that return for shareholders. So that's how we got to having a substantial position. Now that the position is no longer in a private company but in a public company, we have the opportunity to sell and that opportunity comes in tranches.
We've had a few of those and we'll have more over the coming weeks. We're at just over 15% of the portfolio in this holding today. And you know, we will just continue to work to that point about, you know, maximizing the risk-adjusted returns for shareholders.
Yeah. Yeah. That's really interesting. It's great to see it having contributed so strongly and it being that sort of private into the public, that crossover that SM is positioned so well to offer investors. Sticking with SpaceX, Lawrence, I'd love to hear your take on the investment case looking forward because this has been another common question. The valuation today and what underpins your excitement looking ahead.
What I think underpins the excitement is effectively what happens as launch cost continues to fall. Before SpaceX, it costs $18,000 to $19,000 to get a kilogram of mass into orbit. With Falcon 9 today, the list price is about $2,700 to get a kilogram into orbit. With Starship, they reckon that they're getting that to about $900. It's still a work in progress. Starship, they're still testing. It's not yet reusable. As they move to it being reusable, the ambition is to get that to between $100 to $200 per kilogram of mass into orbit. And I think what underpins the excitement of SpaceX is that as you reduce that cost to launch so dramatically, you take businesses that used to be very much in the realm of science fiction and you make them reality. And the first one that they've done that of real great scale has been Starlink, the connectivity business, launching thousands of satellites in a dense network across the earth to provide global connectivity, internet connections. And you can see in the financial results of Starlink what that's achieved. It's a $17 billion annualized revenue run rate. It's growing nearly 70% year-on-year. It's very profitable. It's making operating profit margins nearly at 40%.
And I think the real attraction with SpaceX is that that is one application of lowering launch cost. And as that launch cost continues to fall to the hundreds of dollars, you start to unlock multiple Starlinks. You move into having compute and data centers in orbit, that becomes a possibility when you move into the hundreds of dollars of cost to get into orbit. Defense applications become numerous. Star Shield, for example, in the US with missile defense. As you reduce the cost of launch, you start to move into the ability to do things like microgravity manufacturing. All of these applications open up. And so when you take a step back of what SpaceX is, it is in effect a near monopoly on the access to the rest of the universe. And we think that is incredibly valuable. So are the starting multiples on profits and revenues high today? Yes, they are. But if you're able to build multiple businesses of Starlink scale, then that can generate a good long-term return, we believe, over the next decade for Scottish Mortgage shareholders.
Yeah. And that's where the long-term return really comes in because you can start thinking about what happens if you unlock these physical possibilities and remove constraints of energy and cooling with AI, for example, in space. Incredible what that company has done so far and exciting to see what comes next. Lawrence, we'll stick with you for a moment and we'll shift to another big theme in the questions, which was AI. And Ethan talked about bottlenecks continually moving and different organizations getting different results from the same technology. So, one of the questions you put to Ethan was about the idea of durable value and where that occurs and our audience would like me to put that question to you. So, how do you think about that?
I'd start with something we didn't touch on much with Ethan, which is I think one place where value goes very strongly and will continue to go is to the companies that are enabling intelligence at the chip level, because these companies own bottlenecks within the supply chain. They're in dominant positions and able to develop really, really significant profitability and growth. And those are companies like Nvidia, TSMC, ASML, SK Hynix. And I've talked before about how in many ways these companies are a royalty on the development of AI. Whatever applications become super successful in the next 10 years, whether it's autonomous cars, whether it's robotics, whether it's interesting new software applications, coding agents, shopping agents, these companies benefit. Whether it's on an Anthropic model, Gemini, or OpenAI, again they benefit. And when we talk to all the different AI companies, they all have a different vision of how the world might play out with AI. Open source, closed source, various variations on what AI might look like. But the one thing that they all come back to is that we're going to need a lot more chips.
And so that's one area that's of great interest to us. Now, moving on a little bit more to what we talked about with Ethan about kind of existing organizations and how you capture value there. Where I think what becomes really important is having founder leadership that enables you to be a dynamic and adaptable organization and to pivot. And I think Tobi Lütke at Shopify identifying AI early and focusing the business on that is a key example of that. I think having the right culture that allows that to be possible, being able to attract the best talent. And perhaps a slight difference from Ethan, I do think that there are some companies where having access to proprietary data does make a strong difference because you're able to get more out of that data that no one else can because they don't have that data and use AI to make more use of it. And so that's companies that are doing underwriting better because they're able to use AI models. It's MercadoLibre using all the data it has on transactions and on finance to apply that data to advertising and apply it to making better decisions on underwriting and creditworthiness. So I think there's a combination of things there that allow companies to capture value. The ability to be dynamic and adapt, culture, talent, proprietary data, all of those things really matter.
Yeah. And the presence of founder leadership, you can find that across the portfolio beyond just AI. And it's something about the long-term vision, understanding the supply and the demand and helping companies move so quickly, which is so advantageous in a fast-moving technological space. Tom, we'll stick on AI and I'll come to you please with a question. I mean you framed earlier the sort of the return structure as being one where the upside can so seriously outweigh the downside and so we don't want to emphasize the aspect of risk too much in this discussion, but I think investors are curious about your take on enormous capital flows going into AI. There's this word often used in reporting which is circular capital and I think they're curious to hear about how the balance of risk and reward in AI feeds into your views of portfolio construction.
Yeah, I wouldn't necessarily agree with your suggestion of the question of we shouldn't focus too much on risk. I mean, I think absolutely you need to think about risk and return in investment. I just would go with what you call risk and it's risk is not volatility. Volatility is inevitable. Risk is much more about permanent loss of capital. It's about the impact of things you did not expect. So I wouldn't hide from that aspect of it. On the broader point about how do you think about what's going on in AI, I think before you get into sort of evaluating what's happening in financing and supply chain, I think you have to start with this idea of demand. You know, are people getting value from this technology? Is it, you know, is it, are people going to demand more tokens in the future or fewer? And the trend we see is that people are just using more and more tokens. And the other trend you can observe is if you look at the biggest users of AI, whether that's individuals or companies versus the median, that distance is getting larger as well. So the biggest users of tokens on OpenAI, for example, they've said are about eight times now the median company. So the people that are really leaning into using this technology are seeing greater and greater value from it. I think the coding use case is really illustrative here that you've just seen these huge leaps in productivity in coding. More and more of the computer code that's written is being done by these AI models. Just really clear path to economic value for the companies involved. So you've got a really strong set of demand drivers and I think they only get stronger. You know, we've gone from, you know, models that you talk to to these so-called agentic models that operate on their own. They just consume more and more tokens. So the demand is really clear.
There's, as Lawrence has touched on, therefore there is a huge need for more compute infrastructure and that is extremely expensive. These numbers are enormous. The only context in which they're not enormous is the cash flows of the hyperscale cloud companies. You know, the likes of Google, of Amazon, of Meta. You know, these companies can afford to invest in this kit and they can't afford to miss out on this jump to this new computing paradigm, you know, because that would be so costly for their business in the long run. And so they're throwing all the capital at it that they can and then they're going to debt markets to finance this. They're going to equity markets to finance this. And you're seeing suppliers like Nvidia provide the financing for other players because it's absolutely in their strategic interests for there to be more providers of this infrastructure. So there's just a wall of capital being thrown at this. And I think, you know, we don't know exactly where this plays out. You know, it's very possible this is a volatile ride. But what we do know with a reasonably high degree of confidence is that we're going to need a lot more compute than we can build today in the future if the potential of these systems continues to grow.
Yeah. So as so often, it's the idea that you're avoiding the market's preoccupation with the idea of volatility as a measure of risk, and it's understanding what the actual underlying risk is in relation to the rewards that you see the potential of.
So, thank you for that. Lawrence, just one more question on AI before we move on to topics more broadly. One of the things that came up in the discussion with Ethan and we've heard a little bit about it here is this idea of return on investment and it's a question that's come in time and time again is, is it about waiting for all of your cash flows in the future or what are you hearing in terms of signal from cash being generated today? How are the companies you're speaking to reporting on that?
The way I tend to frame this is 18 to 24 months ago when we were meeting different companies including many in the portfolio and others we don't own, companies were experimenting with AI. They were trying it in different ways and seeing how it could be useful. I think over the last sort of 12 months that's really shifted where now what they're telling you is we think this is going to have a really material impact on our business. It's going to allow us to make better products. It's going to allow us to accelerate revenue growth or reduce cost in some way. And I'd go a step further that some of the more recent conversations I've had with the heads of technology of some of our holdings have been actually starting to say we need to think about how we rearchitect our entire organization around AI models and how we work with them. We need to not just change our headcount but change the type of people and the roles and some people in organizations will adapt to this and others won't. And so you can see as these conversations have gone on with companies, the impact is getting bigger and bigger. And what almost all of them say is we're going to be using more tokens in the future. So more compute, more intelligence. And we believe that we are getting a positive return on that investment. So we will keep investing behind it. And then I think of it in terms of my own use of AI. I spend more time for my work using AI every month than I did the month before.
That's my time increasing, but then because the models are more sophisticated and they can do more sophisticated things, my tokens are increasing much faster than my time spent is. So my token usage is increasing exponentially. And when I look at what that costs, I know that I'm getting a really good return on investment. And when I speak to the software CEOs of companies that are using AI within their business, they also say actually we're getting a really good return on investment and we worry about, you know, what happens if the pricing goes up because we're getting such a good return it might invite pricing increases. So across all of these different data points we're seeing is that at the moment people are finding a return on that investment.
